research.md

back to table · edit · history · where entries came from · files · download

4666 bytes, as of the version from 2026-11-17 07:50 (current). Recorded here, not run.

# T734 build research

Checked: the archived local/live canonical bundles match at bundle SHA-256
984965681e2f8925c511a121c118efcde59581c7b0d906cac8b641bf6925e55a.
Manifest: ../20261117T072043Z-build-skills/manifest.json. All resource file
hashes were revalidated. The local core and six task-relevant references were read.

Checked source passages: AM arXiv:2402.14216v1, introduction and (1.1) define
c0 and g with a minus cotangent sum, positive reduced denominator and empty
sum zero. Its section 3, (3.1)–(3.2), provides the upward contour and functional
equation. BC, Period functions and cotangent sums, printed pp. 219 (3),(4),
223–224 (12) and 226 (14), defines E1, psi0 and the reciprocity normalization.
Multiplying BC (14) by x gives g(x)=i*x*psi0(x)/2. AM's introductory printed
psi=x*g/(2*i) contradicts those defining equations; at x=1 it is off by four.
This is a localized scale error, not an unresolved definition. Do not identify
BC's auxiliary g0 with AM's g. Natural logs, positive pi, x>0 are used here.
PDF/HTML downloads and hashes are listed in sources.json; extracted passages
are bc-passage-page-{6,7,10,13}.txt and am-text.txt. No source decimal is imported.

Checked live discovery: catalogue.json caches one complete authenticated pass
(730 records), filtered into candidates.json; refreshed metadata-at-claim files
show T639 (c0), T692 (normalized Taylor coefficients), T693 (signed asymptotic
coefficients), T710 (ingredient minimal polynomials), T640 (fractional-part
integral). None stores g values. Their definitions and coverage were read.
These tables are companion context, not independent numerical evidence.
No matching draft was listed. already_here and already_asked results are in
screening.json: #330 is the active proposal, #308 is the earlier coefficient
family. Numeric /api/lookup checks at five expected values returned no matches.
The first malformed lookup attempts and their errors are retained. Existing
number theory, special values and integral tags were checked on two live pages;
quantum modular forms was absent. A recommendation could connect T639,T692,T693
and T734, but no new tag or changes to other tables are made here.

Checked provenance: proposal.json maps only contributes to #273. Live #273 and
its discussion concern approximation/reciprocity; #275/#276 are broader context,
not direct origins of this table. Initial message cites #330 and #273 explicitly.
claim.json records private draft T734; no offered/public status was requested.

Mathematical decisions: one positive-real parameter x with exact reduced
rational identities. Full coprime numerator/denominator coverage through 20 is
arithmetic structure selected by the rational law, not arbitrary independent
couplings. The bound is noncanonical. selection-plan.json and argument-selection.md
retain every identity and the intrinsic x=1 / arithmetic-structure partition.
Irrational and complex coverage is deferred rather than sampled mechanically.

Checked derivations: contour-derivation.md derives the explicit tail majorant
using inspected AM and DLMF formulas, and proves reciprocal/three-term identities
from the finite definition. Production has no truncation: it evaluates finite
cotangent sums in Arb with exact rational inputs. A separately implemented finite
Estermann sum and a 15-point analytic-contour sample provide build checks.
Actual results, controls, radii and coverage are recorded in the result JSONs;
this document does not stand in for an independent scientific verdict.

Research question, not theorem: for 0<x<=1, compare
R(x)=pi*g(x)+log(2*pi*x)-gamma with (pi^2/36)*x^2. Broad and exact-term web
searches did not locate this precise inequality; that is not a novelty claim
or an exhaustive literature review. The source only supplies the contour used
for the experiment. Development denominators through 12 and withheld 13–20
are recorded separately by check_values.py. Neither proposed positivity nor
an RH assumption enters the evaluator. All counterexamples, if any, are kept.

Completed build measurements: production minimum 445 relative bits, independent
Estermann minimum 534, and the 15 analytic contour points minimum 358. The
contour tail is below 2e-133 and the maximum final contour radius is below
1.08e-110. All 255 actual stored decimal intervals were checked independently;
255/255 also reproduce with the generator. The 128-point inequality experiment
has no counterexample (46 development, 82 withheld), without a universal proof.
Audit JSON retains two advisory findings; audit-assessment.md explains the
arithmetic-coverage and defining-cross-reference judgments for the fresh critic.